Plotting

Plotting basics

For the R and Python sections here, either pick one of the languages, or do both.

For R, do both base and ggplot2 operations!

Note that I’m not having you do multi-panel plots in base R.
Spoiler alert: it is kinda annoying.

R

Greek symbols

Take the following plot:

plot(sin)

Change it so that the x axis label is \(\sigma\) and the y axis label is \(\tau\). Those are the Greek letters “sigma” and “tau”, respectively.

Hint: look up the R function expression.

Take the same plot, and change the x axis label to \(\sigma^2_s\).

Color and line style (base graphics)

For the same plot as above, change the plot so that the line is red, “dot-dash”, and 3 times the default line thickness.

Axis labels from data

Take the graph we started with, and assume there’s a variable y = 7. Add a title to the graph that says \(\Gamma = 7\). (That’s the Greek letter “gamma”.) Do it using the variable! Don’t “hard-code” in the value!

Hint: the function you need here is bquote and the horrific .(foo) syntax, where foo is a variable. Extra hint: this is one of the more obnoxious corners of R.

Annotations from data

Take the graph we started with, and assume there’s a variable y = 7. Add a text element at coordinates x=0.1, y=0.4 to the graph that says \(\Gamma = 7\).

Saving files

Save one of your graphics in the following formats:

pdf with 18 point font, 10" by 10" square
tiff with 18 point font, 10" by 10" square and the x axis font is 1/2 the global font size.
72 dpi png with 18 point font, 10" by 10" square
300 dpi png with 18 point font, 10" by 10" square

Note: please do not commit the files! Some of them will be massive! The point is for you to learn the commands.

Hints for R users:

See help(par)
Each of the above outputs will require at least 4 line of R code.

R/ggplot

Take the following ggplot:

library(ggplot2)
data(mtcars)
p = ggplot(mtcars, aes(x=cyl, y = mpg)) + geom_point()
p

Repeat the various manipulations above. Where the above instructions refer to lines, change point color/size/shape instead. New users: ggplot does not use par, cex, etc.. It has its own “special” ways.

Also, use the ggpubr library to plot two ggplots side by side. (It is fine to plot the same plot next to itself.)

Do it again, but as a 2x2 matrix of 4 plots.

Python

Do the same manipulations described above for the R plots, but using this graph:

Hint: all of the Greek letter stuff can be accomplished with string literals and \(\LaTeX\) syntax. (This is an area where Python is far easier than R.)

For the ggplot parts, use seaborn.

For multi-panel plotting, you may use matplotlib.pyplot or seaborn as you see fit. Hint: you’ll want to take a look at GridSpec, which is part of matplotlib.

import matplotlib.pyplot as plt

f, ax = plt.subplots()
x = [i for i in range(10)]
y = [i**2 for i in x]
ax.plot(x, y)
ax.set_xlabel('x')
ax.set_ylabel('y')

Questions from the group

Narges

Heat maps with marginals in Python.

Sometimes, a “heat map” is confused with a joint distribution.

Here’s a joint distribution:

import seaborn as sns
import matplotlib
print(sns.__version__)

## 0.11.1

df = sns.load_dataset('iris')
sns.jointplot(x=df["sepal_length"], y=df["sepal_width"], kind='hist', palette='cubehelix', hue=df['species']);
plt.show()

And a heat map:

import numpy as np
import pandas as pd
np.random.seed(42)

data = np.random.multivariate_normal([0., -0.25], [[1., 0.0], [0.0, 4.]], 10000)
print(data)

## [[-0.1382643   0.74342831]
##  [ 1.52302986  1.04537708]
##  [-0.23413696 -0.71830675]
##  ...
##  [ 0.38782734  1.65518102]
##  [ 0.37835397  2.50473316]
##  [-1.6199198   3.17705946]]

sns.heatmap(data, cmap="viridis")

The joint density version of the above is:

x, y = data.T
df = pd.DataFrame({'x': x,'y': y})
sns.jointplot(x=df['x'], y=df['y'], kind='kde');
plt.show()

Or,

sns.set_palette('viridis')
sns.jointplot(x=df['x'], y=df['y'], kind='hist');
plt.show()

Or, super-sneaky:

from mpl_toolkits.axes_grid1.inset_locator import inset_axes

ax = sns.jointplot(x=df['x'], y=df['y'], kind='hist');
# replace main plot with plt.hist2d
ax.ax_joint.cla()
plt.sca(ax.ax_joint)
h = plt.hist2d(df.x, df.y, bins=(50, 50), cmap='viridis')

# Add color bar onto inset axis
cbaxes = inset_axes(ax.ax_joint, width="45%", height="3%", loc=1)
cbaxes.axis['bottom'].major_ticklabels.set_color('white')
cbar = plt.colorbar(h[3], cax=cbaxes, orientation='horizontal')
cbar.outline.set_edgecolor('white')
plt.show()

Parvan - Plotting from summary stats

This is hard, and not in the usual domain of statistical software.

An incomplete R solution follows. I couldn’t be bothered to get the x-axis spacing to work out, as I can already tell it won’t generalize to a plot with more columns.

Get the raw data and just plot them, I guess.

a = rnorm(1e3, 1, 0.1)
b = rnorm(1e3, 2, 0.5)
heights = c(mean(a), mean(b))
std_dev = c(sd(a), sd(b))
xlim=c(0, length(heights))
barplot(heights, xlim=xlim, ylim=c(0, max(heights) + max(std_dev)))

# Manually add +/- 1 sd
segments(0.5, heights[1], 0.5, heights[1] + std_dev[1])
segments(0.5, heights[1], 0.5, heights[1] - std_dev[1])
segments(1.5, heights[2], 1.5, heights[2] + std_dev[2])
segments(1.5, heights[2], 1.5, heights[2] - std_dev[2])

Johnny - plots aren’t as nice as Graph Pad Prism:

Honestly, I didn’t see anything in that package that couldn’t be done in R. In fact, I almost suspect that they use R or Python as a back end. I think it is just a question of adjusting styles.

Super special hints for Python users

Use \(\LaTeX\) for font rendering.

Some folks don’t like the default matplotlib fonts. Use the beautiful Computer Modern font instead.

Warning: requires a full \(\LaTeX\) setup! I had to do the following first on my Ubuntu:

sudo apt install cm-super dvipng

The code:

import matplotlib.pyplot as plt
from matplotlib import rc
rc('font',**{'family':'sans-serif','sans-serif':['Helvetica']})
## for Palatino and other serif fonts use:
#rc('font',**{'family':'serif','serif':['Palatino']})
rc('text', usetex=True)

f, ax = plt.subplots()
x = [i for i in range(10)]
y = [i**2 for i in x]
ax.plot(x, y)
ax.set_xlabel('x')
ax.set_ylabel('y')

Remote plotting

On Unix systems, the default display back-end for matplotlib requires X11. This requirement poses a problem when trying to have an HPC job generate a plot–a user authenticated to use an X display does not exist. Likewise, it is a problem when working remotely via ssh and you didn’t use -Y to connect, etc..

Solution: switch the back-end to agg:

# Not executed in this notebook!
import matplotlib
matplotlib.use('agg')